The efficacy of digital cognitive behavioral therapy for insomnia and depression: a systematic review and meta-analysis of randomized controlled trials

Search strategy

The current study was carried out according to the Preferred Reporting Items for Systemic Reviews and Meta-Analysis criteria (PRISMA) guidelines (Page et al., 2021). The study protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO, http://www.crd.york.ac.uk/prospero), registration number: CRD42022355049. The methodology differs from the registered protocol on PROSPERO, because adjustments have been made during the review process. Between-group comparisons for depression symptoms (registered on PROSPERO) at post-treatment and follow-up time points were conducted, while the differences in insomnia severity at post-treatment and follow-up time points were calculated. The search strategy was designed by two researchers (Wenyao Lin and Na Li) using a combination of free words and relevant Mesh terms (dCBT-I, insomnia, and depression). The search strategy is detailed in Table S1. PubMed, Cochrane, Embase, and Web of Science databases were searched from 1 January 2000 to 5 June 2022. Grey literature, existing systematic reviews, and meta-analysis were also searched to identify potential studies which met the inclusion and exclusion criteria. Furthermore, we updated the search to collect relevant publications during 2022 and found two additional relevant articles (Wiklund et al., 2022; Mason et al., 2023). Due to the small amount of updated literature, we did not update the data analysis.

Inclusion and exclusion criteria

The inclusion and exclusion criteria for primary studies were designed to select high-quality clinical trials, allowing for drawing reliable conclusions regarding the efficacy of dCBT-I in patients with insomnia and depression. The inclusion criteria followed the Participant, Intervention, Comparison, Outcome, and Study (PICOS) guidelines (Liberati et al., 2009). Studies meeting the following criteria were included in this meta-analysis:

Data extraction

Data were independently extracted from the included studies by two researchers (Wenyao Lin and Lili Yang). Any difference in the extracted data was resolved by discussion. After literature selection, the following information was manually extracted from all included studies: the first author, year of publication, sample size, age of participants, percentage of females, therapeutic components, treatment duration, diagnostic approach for insomnia and depression, mean and standard deviation for the severity of insomnia and depression at baseline, post-treatment, and final follow-up.

Quality assessment

The methodological quality of the included studies was assessed using the Cochrane Collaboration risk of bias assessment tool (Higgins et al., 2011). Each study was evaluated from seven specific domains: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other sources of bias (sample size estimation, complete definition of intervention sessions and outcomes in the intervention and control groups). Two researchers independently rated each domain as low (green), unclear (yellow), or high (red) risk of bias. If there were any dissents, the corresponding author was consulted.

Statistical analysis

Statistical analyses were performed using Revan Manager (version 5.4). Outcomes of insomnia and depression were all continuous variables, and therefore standardized mean difference (SMD) was used to pool effect sizes. Effect sizes were calculated by subtracting the mean post-treatment/final follow-up score in the treatment group from the mean score in the control group, and dividing the result by the pooled standard deviation of the two groups. Using the post-treatment/follow-up mean score, instead of a change in mean scores between baseline and post-treatment/follow-up, is less likely to cause bias in effect sizes, and baseline differences do not necessarily lead to discrepancy in conclusions (Fu & Holmer, 2016). Effect sizes of 0.2, 0.5, and 0.8 are considered to be small, medium, and large respectively (Borenstein et al., 2010). Assuming that there are multiple true effect sizes rather than a single true effect size, a random-effects model (Borenstein et al., 2010) was adopted for data analysis. Statistical heterogeneity among included studies was evaluated with I² statistics. The I² values of 0% to 40%, 41% to 74%, and 75% to 100% represent low, medium, and considerable heterogeneity, respectively (Higgins & Green SJN-SAfePuP, 2011). All tests were two-sided, with p < 0.05 considered statistically significant.

Sensitivity analysis and publication bias

Sensitivity analysis was performed to explore the source of heterogeneity. Publication bias was visually assessed using funnel plots and statistically examined using Egger’s tests (Egger, Smith & Phillips, 1997). The sensitivity analysis and publication bias test were performed using the STATA statistical software (version 15.0; College Station, TX, USA).

Retrieval results and study characteristics

A total of 207 records were identified in the database search. After duplicated records were removed, there were 162 records left. Upon the title and abstract screening, 83 ineligible records were excluded. Then, after the full texts of the remaining 119 records were read, 83 records that were not meet the inclusion criteria were deleted. Of the remaining 36 studies, seven articles were included in the present study. The literature screening process is shown in Fig. 1.

Of the included seven studies (Horsch et al., 2017; Cheng et al., 2019; Freeman et al., 2017; Rajabi Majd et al., 2020; Kyle et al., 2020; Faaland et al., 2022; Kalmbach et al., 2020), most were conducted in European regions, including UK (n = 2), Norway (n = 1) and Netherlands (n = 1). The remaining studies were carried out in USA (n = 2) or China (n = 1). The dCBT-I was delivered via mobile phone apps in three studies (n = 3) and via internet programs in other studies (n = 4). The diagnostic criteria for insomnia were based on the DSM-5 in two studies (Cheng et al., 2019; Kyle et al., 2020), the Insomnia Severity Index (ISI) in three studies (Horsch et al., 2017; Faaland et al., 2022; Kalmbach et al., 2020), both in one study (Rajabi Majd et al., 2020), and the Sleep Condition Indicator in one study (Freeman et al., 2017). The dCBT-I employed in all included studies was multi-component, including sleep hygiene education (SHE), cognitive restructuring (CR), sleep restriction (SC), and stimulus control (SC). The dCBT-I components differed between studies. The specific dCBT-I components in the included studies are shown in Table 1. In contrast, the program length was mostly similar. Six studies provided a 6-session dCBT-I program, while one study adjusted the dCBT-I program for each participant (Horsch et al., 2017). Most of the included studies were done within 12 weeks (1, 9, 10 or 12 weeks), while one study examined the effect of dCBT-I after 3 months of treatment (Horsch et al., 2017). Four studies reported the follow-up time, at 6 weeks (n = 1), 22 weeks (n = 1), and 24 weeks (n = 1), 6 months (n = 1), respectively. All included studies used the ISI to assess the severity of insomnia. At the same time, depression was measured using different scales, including Quick Inventory of Depression (QIDS) (Cheng et al., 2019), Hospital Anxiety and Depression scale (HADS) (Rajabi Majd et al., 2020; Faal et al., 2022), Patient Health Questionnaire (PHQ) (Freeman et al., 2017; Kyle et al., 2020), Edinburgh Postnatal Depression scale (EPDS) (Kalmbach et al., 2020), and Center of Epidemiological Studies Depression scale (CES-D) (Horsch et al., 2017).

There were 3,597 participants in the dCBT-I treatment group (mean age = 24.8–52.5, 1990 (55%) females) and 3,500 in the control group (mean age = 24.6–52.4, 1928 (55%) females). One study did not report age or gender. The smallest sample size was 101, and the largest was 3,755. The dropout rate in each study was calculated, with an average of 29.8% in the intervention group and 23.6% on the waiting list. Table 2 shows the basic characteristics of the included studies.

Quality assessment

The Cochrane Collaboration risk of bias assessment tool was used to assess the quality of the included studies. Most studies used random sequence generation, had no selective reporting, and completely defined intervention sessions and outcomes in both intervention and control groups. Allocation concealment and blinding of participants and personnel were performed in most of the studies. More than half of the included studies did not report whether the outcome assessment was blinding. The quality assessment of the included studies is presented in Figs. S1 and S2.

Short-term and long-term effects of dCBT-I on insomnia

The self-reported ISI was provided in all eligible studies (Gullickson & Ramser, 1996), involving 5,779 participants (n = 2911 in the treatment group and n = 2868 in the control group). The effect size (Hedges’g) at post-treatment was calculated to quantify the short-term effect of dCBT-I on insomnia. The meta-analysis showed that dCBT-I significantly improved insomnia at post-treatment (SMD = −0.85, 95% CI [−1.00 to −0.69], p < 0.001) (Fig. 2).

Four studies conducted follow-up evaluations and therefore included in the meta-analysis of the long-term effect of dCBT-I on insomnia. In total, 932 participants in the treatment group and 1,330 in the control group completed the final follow-up evaluations (ranging from six weeks to six months). The dCBT-I also showed a significant improvement in insomnia at final follow-up (SMD = −0.71, 95% CI [−1.00 to −0.44], p < 0.001). More details of the effect of dCBT-I on insomnia at final follow-up are shown in Fig. 3.

Short-term and long-term effects of dCBT-I on depression

As depression was measured using different scales, only included four studies were included: two studies measured depression with PHQ (Freeman et al., 2017; Kyle et al., 2020) and two studies measured depression with HADS (Rajabi Majd et al., 2020; Faal et al., 2022). Other three studies used QIDS (Cheng et al., 2019), EPDS (Kalmbach et al., 2020), and CES-D (Horsch et al., 2017) respectively. The results of meta-analyses showed that dCBT-I significantly reduced the PHQ score compared to the control group (SMD = −0.47, 95% CI [−0.55 to −0.38]), p < 0.001). However, there was no significant difference in the HADS score between the dCBT-I treatment group and the control group (SMD = −0.57, 95% CI [−1.33 to 0.19], p < 0.001) (Fig. 4).

The effect of dCBT-I on depression at final follow-up is shown in Fig. 5. dCBT-I also resulted in a statistically significant decrease in the PHQ score at the final follow-up (SMD = −0.42, 95% CI [−0.68 to −0.15], p = 0.002).

Sensitivity analysis and publication bias

Sensitivity analysis showed that the meta-analysis results were relatively stable for insomnia (Fig. S3). Despite high heterogeneity in the ISI for the post-treatment (I² = 79%) and final follow-up (I² = 84%), Egger’s test showed no publication bias (post-treatment, p = 0.449; final follow-up, p = 0.974). The funnel plot for the ISI was nearly symmetrical, as presented in Fig. S4. As there were only 2 RCTs included in the meta-analysis of PHQ score and HADS score to evaluate the effect of dCBT-I on depression, respectively, we could not make an inverted funnel plot to assess the influence of publication bias on included studies.